MIN Faculty Department of Informatics Coordination in Multi Robot Systems Motion Planning Shreyans Kumar Bhansali University of Hamburg Faculty of Mathematics, Informatics and Natural Sciences Department of Informatics Technical Aspects of Multimodal Systems 18. December 2017 A. Studentr – Presentation Title 1 / 29
Outline Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion 1. Motivation 2. Motion Planning 3. Centralized Approach 4. Decentralized Approach 5. Conclusion A. Studentr – Presentation Title 2 / 29
Motivation Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion 1 1 <https://www.popsci.com/these-new-motivational-posters-will-get-your-robot-out-its-slump> A. Studentr – Presentation Title 3 / 29
Motivation Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Why do we need coordination in multi-robot systems? ◮ Force multiplication (PRL-based research in multi-rover coordination for conceptual lunar-surface assembly operations. 2 ) 2 <https://www-robotics.jpl.nasa.gov/facilities/facility.cfm?Facility=4> A. Studentr – Presentation Title 4 / 29
Motivation Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion ◮ Simultaneous presence (Multi-robot system in an eye-in-hand configuration. 3 ) 3 <http://www.mdpi.com/1424-8220/11/10/9839/htm> A. Studentr – Presentation Title 5 / 29
Motivation Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion ◮ Faster execution ◮ Redundancy, fault tolerance ◮ Greater efficiency ◮ Larger range of task domains - cooperative manipulation (Cooperative manipulation 4 ) 4 <https://homepages.laas.fr/afranchi/robotics/?q=node/251> A. Studentr – Presentation Title 6 / 29
Applications Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion ◮ Warehouse management ◮ Competitions - Robot Soccer ◮ Product assembly ◮ Digital entertainment 5 6 7 (a) Kiva Systems/Amazon (b) Robot Soccer (c) Clash of Clans Example Applications 5 <https://www.youtube.com/watch?v=lTJ1sIqBoro> 6 <https://tams.informatik.uni-hamburg.de/lehre/2016ws/seminar/ir/doc/slides/JuliusMayer- Genetic_Algorithms_in_Robotics.pdf> 7 <https://www.youtube.com/watch?v=Bzc7P99be9E> A. Studentr – Presentation Title 7 / 29
Research Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion ◮ Multi-robot motion planning ◮ Traffic control ◮ Multi-robot docking ◮ Foraging ◮ Multi-robot soccer ◮ Exploration and localization A. Studentr – Presentation Title 8 / 29
Research Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion ◮ Multi-robot motion planning ◮ Traffic control ◮ Multi-robot docking ◮ Foraging ◮ Multi-robot soccer ◮ Exploration and localization A. Studentr – Presentation Title 9 / 29
Objective Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Enable robots to navigate collaboratively to achieve spatial positioning goals. [6] Motion planning in dynamic environment with moving obstacles is NP-hard A. Studentr – Presentation Title 10 / 29
Approaches Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion (Motion planning in Multi-robot systems 8 ) ◮ What information does an approach access? ◮ Global = Centralized ◮ Local = Decentralized 8 http://multirobotsystems.org/sites/default/files/slides/2015_RSS_MRS_Bekris.pdf A. Studentr – Presentation Title 11 / 29
Approaches Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion (Motion planning in Multi-robot systems 9 ) Key question for Centralized Approaches ◮ What space does the approach search over? ◮ Composite space of all robots = Coupled approach ◮ Individual robot space and co-ordination = Decoupled approach 9 http://multirobotsystems.org/sites/default/files/slides/2015_RSS_MRS_Bekris.pdf A. Studentr – Presentation Title 12 / 29
Approaches Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion (Motion planning in Multi-robot systems 10 ) Key question for Decentralized Approaches ◮ How does a local method access information from robots? ◮ Sensing or communication ◮ Inference or shared information 10 http://multirobotsystems.org/sites/default/files/slides/2015_RSS_MRS_Bekris.pdf A. Studentr – Presentation Title 13 / 29
Centralized Approach - Coupled Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Treat multiple robot as just one robot Single agent "leader" plans for the entire team ◮ Plan path in configuration space Q = Q1 x Q2 x . . . QN (Centralized approach 11 ) 11 <http://www.robotmotionplanning.org/teaching/LecRoboMultiRobots.pdf> A. Studentr – Presentation Title 14 / 29
Centralized Approach - Coupled Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Treat multiple robot as just one robot Click! 12 12 <https://www.cs.rutgers.edu/ kb572/pubs/scalable_asympt_opt_multi_robot.pdf> A. Studentr – Presentation Title 15 / 29
Centralized Approach - Decoupled Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion ◮ First compute individual path for each robot for their corresponding space Qi ◮ The consider path interaction to produce a solution in composite space ◮ When successful, they solve problems faster than coupled approach Various methods for Decoupled approach are: ◮ Prioritized planning ◮ Velocity tuning A. Studentr – Presentation Title 16 / 29
Centralized Approach - Decoupled Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion (Decoupled approach 13 ) 13 <http://www.robotmotionplanning.org/teaching/LecRoboMultiRobots.pdf> A. Studentr – Presentation Title 17 / 29
Centralized Approach Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Advantages: ◮ Single robot motion planning algorithms can be directly applied ◮ Leader can take all information into account ◮ In theory, co-ordination can be perfect ◮ Guarantees probabilistic completeness Disadvantages: ◮ Computationally hard ◮ Vulnerable to malfunction of leader ◮ Heavy communication load A. Studentr – Presentation Title 18 / 29
Decentralized Approach Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion ◮ Control processing is distributed among agents ◮ Each robot basically independent ◮ Robots use locally observable information to make plans (Decentralized approach 14 ) 14 <http://www.robotmotionplanning.org/teaching/LecRoboMultiRobots.pdf> A. Studentr – Presentation Title 19 / 29
Decentralized Approach Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Why do need decentralized approach? A. Studentr – Presentation Title 20 / 29
Decentralized Approach Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Why do need decentralized approach? Robust to limits on or loss of communication A. Studentr – Presentation Title 20 / 29
Decentralized Approach Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion (Multiple robot systems using Dynamic Networks 15 ) 15 http://multirobotsystems.org/sites/default/files/slides/2015_RSS_MRS_Bekris.pdf A. Studentr – Presentation Title 21 / 29
Decentralized Approach Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion (Multiple robot systems using Dynamic Networks 16 ) 16 http://multirobotsystems.org/sites/default/files/slides/2015_RSS_MRS_Bekris.pdf A. Studentr – Presentation Title 22 / 29
Decentralized Approach Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion ◮ Every time a new network is formed, data is exchanged. ◮ Each robot uses its own centralized motion planner to construct trajectories ◮ After each robot has received a plan from all other robots, it will implement the best plan. A. Studentr – Presentation Title 23 / 29
Inter-robot communication Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Objective of communication A. Studentr – Presentation Title 24 / 29
Inter-robot communication Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Objective of communication Enable robots to exchange state and environmental information with a minimum bandwidth requirement [4] A. Studentr – Presentation Title 24 / 29
Decentralized Approach Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion Advantages: ◮ Dimensionality of configuration space does not increase ◮ Faster response to dynamic conditions ◮ Little computation required ◮ Very robust Disadvantages: ◮ Plans based only on local information ◮ The solutions are often highly sub optimal A. Studentr – Presentation Title 25 / 29
Conclusion Motivation Motion Planning Centralized Approach Decentralized Approach Conclusion A. Studentr – Presentation Title 26 / 29
Recommend
More recommend
